spark-ds41f-1025: DeepSeek V4.1 Flash inference engine update for DGX Spark
All kernels rewritten in TileLang with bit-identical outputs; prefill up 12–14% on four Sparks, 75 tok/s single stream, two logprob bugs fixed, 5,730 tests passing.
spark-ds41f-1025 is out. DeepSeek V4.1 Flash on DGX Sparks. 1M context. Native weights. 1. Every kernel the server runs is TileLang now. B12X only loads the weights. 2. Outputs: same as last release. Bit for bit. Logprobs too. 3. Prefill: +12-14% on four Sparks. +23-28% on three. 4. Decode steps 1.7-2.4% shorter. Throughput: level. 75 tok/s single stream on four. 5. Two logprob bugs fixed. One of them could take the whole engine down. 6. 5,730 tests passed. 0 failed. 7. weightless abliteration (GLP steering). It edits the activations, not the weights. Off by default. Auditable. Won't hack you. 8. Everything is called spark-ds41f now. Almost. Rewrote every kernel. Changed zero bits. That was the point. http://github.com/christopherowen/spark-ds41f